Detect unknown MIME types without copying uploads

[CyMule]

Summary

• detect unknown MIME types directly from the existing spooled upload stream
• preserve the uploaded filename through a lightweight file proxy
• rewind the original upload after detection
• add a regression test that verifies the original stream is used

Why

The previous path read the complete upload into a new BytesIO object before detection. That creates a document-sized allocation per request and can amplify memory pressure under concurrency.

Impact

Detection behavior is preserved while avoiding the full-body copy. A local 40 MiB spooled-upload check measured about 0.065 MiB of peak traced allocation in this path, with the stream rewound to position zero afterward.

Validation

• uv run --locked --no-sync pytest -q test_general/api/test_filetypes.py — 1 passed
• focused Ruff check and format check passed for all changed Python files
• 40 MiB local allocation check — mime=text/plain peak_mib=0.065 position=0

[cubic-dev-ai]

All reported issues were addressed across 4 files

<sub>Shadow auto-approve: would not auto-approve because issues were found.

Re-trigger cubic</sub>

[cubic-dev-ai]

0 issues found across 2 files (changes from recent commits).

<sub>Shadow auto-approve: would auto-approve. Optimizes MIME detection to avoid a full-body BytesIO copy by proxying the spooled upload, preserving behavior and filename. Includes regression tests verifying the original stream is used and rewound. A bounded, performance-improving refactor with no observable external change.

Re-trigger cubic</sub>